Large language models<\/a> answer questions using the knowledge they learned during training. This fixed knowledge base limits them. They can\u2019t give you current or highly specific information. Retrieval-Augmented Generation<\/a> (RAG) helps by letting LLMs pull in external data, but even RAG needs help with complex questions. Adaptive RAG offers a solution. It picks the best approach for answering a question. This could be a direct answer from the LLM, a single data lookup, or multiple data lookups or even a web search. A classifier looks at the question\u2019s complexity and decides the best method. <\/p>\n Here are the three answering strategies Adaptive RAG uses:<\/p>\n Adaptive-RAG is flexible. It handles each question efficiently, using only the resources needed. This saves computing power and makes users happier because they get precise answers quickly.\u00a0 Adaptive-RAG also updates itself based on question complexity. Because it adapts, it reduces the chance of old or too-general answers. It gives current, specific and reliable answers.<\/p>\n Improving Accuracy<\/strong><\/p>\n Adaptive-RAG is very accurate. It understands the complexity of each question and adapts. By matching the retrieval method to the question, Adaptive-RAG makes sure answers are relevant. It also uses the newest available information. This adaptability greatly reduces the chance of getting old or generic answers. Adaptive-RAG sets a new standard for accurate automated question answering.<\/p>\n The image from the research paper compares different RAG approaches. Adaptive RAG excels in both speed and performance.<\/p>\n Adaptive Retrieval-Augmented Generation (Adaptive RAG)<\/strong> is a RAG architecture designed to improve how large language models (LLMs) handle queries. It adjusts the strategy for answering questions based on their complexity.<\/p>\n Adaptive RAG selects the best approach for answering a query, whether it\u2019s simple or complex. A classifier evaluates the query and decides whether to use a straightforward method, a single-step retrieval, or a multi-step process.<\/p>\n As shown in the above image, here\u2019s the comparison of adaptive RAG with a single-step, multi-step approach.<\/p>\n This method retrieves information in one go and pass it to the LLm to generate an answer directly. Although it works well for simple questions like \u201cWhat is Agentic RAG?\u201d, it struggles with complex queries. For instance \u201cWhat is the use of Agentic RAG in the healthcare sector?\u201d due to lack of reasoning on the retrieved information.<\/p>\n This method processes all queries through multiple retrieval steps, refining answers iteratively. While effective for complex queries, it can waste resources on simple questions. For instance, repeatedly processing \u201cWhat is Agentic RAG?\u201d is inefficient.<\/p>\n A classifier determines the complexity of the query and selects the appropriate strategy:\u00a0<\/p>\n Here\u2019s another example of an adaptive RAG System architecture flow from LangGraph:<\/p>\n Adaptive RAG System Architecture Flow:<\/p>\n The process begins with analyzing the user query to determine the most appropriate pathway for retrieving and generating the answer.<\/p>\n If the query is routed through the RAG module<\/strong>, it undergoes an iterative, self-reflective process to ensure high-quality and accurate responses.<\/p>\n If the query is deemed unrelated to the indexed knowledge base during the Query Analysis stage:<\/p>\n In summary, Adaptive RAG is a smart framework that enhances response quality and efficiency by adapting to the complexity of queries. Now let\u2019s build a RAG application using the Adaptive-RAG strategy. Our application sends user questions to the best data source. Here\u2019s the setup:<\/p>\n Here\u2019s a breakdown of how we build the application using the LangGraph framework:<\/p>\n We will be implementing the Adaptive RAG System we have discussed so far using LangGraph<\/a>. We will be loading some of our blog contents using PyPDFLoader into our vector database, the ChromaDB<\/a> vector database. and also using the Tavily Search tool<\/a> for web search. Connections to LLMs and prompting will be made with LangChain<\/a>, and the agent will be built using LangGraph. For our LLM, we will be using ChatGPT GPT-4o<\/a>, which is a powerful LLM that has native support for tool calling. Although, we can use any other LLM, it is advisable to use LLM finetuned for tool calling.<\/p>\n This image explains the workflow of the adaptive rag system.<\/p>\n OPENAI_KEY = getpass('Enter Open AI API Key: ')<\/p> TAVILY_API_KEY = getpass('Enter Tavily Search API Key: ')<\/p><\/code><\/pre>\n We have downloaded the blogs PDFs and made the pdfs documents available in a folder named \u201cpdf\u201d on Google Drive, you can download it from here.<\/p>\n\n
Why Adaptive-RAG Works So Well?<\/h2>\n
![\"QA](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/image-105-1.webp\")
What is Adaptive RAG?<\/h2>\n
The Key Idea of Adaptive RAG<\/h3>\n
Comparison with other approaches<\/h3>\n
1. Single-Step Approach<\/h4>\n
\n
2. Multi-Step Approach<\/h4>\n
\n
3. Adaptive Approach<\/h4>\n
\n
\n
Adaptive RAG Framework<\/h2>\n
\n
RAG System Architecture Flow from LangGraph<\/h2>\n
1. Query Analysis<\/h3>\n
\n
\n
2. RAG + Self-Reflection<\/h3>\n
\n
\n
\n
\n
\n
\n
\n
\n
\n
3. Web Search for Unrelated Queries<\/h3>\n
\n
Implementation Steps<\/h2>\n
\n
Building an Adaptive RAG System with Self-Reflection<\/h2>\n
Workflow of Adaptive RAG<\/h3>\n
![\"Workflow](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/Workflow-of-Adaptive-RAG.webp\")
Install OpenAI, ChromaDB, LangGraph and LangChain Dependencies<\/h3>\n
!pip install -U langchain_community tiktoken langchain-openai langchainhub langchain_chroma langchain langgraph\u00a0 tavily-python pypdf<\/code><\/pre>\nSet up OpenAI and Tavily API<\/h4>\n
from getpass import getpassSetting up environment variables<\/h4>\n
import os\nos.environ['OPENAI_API_KEY'] = OPENAI_KEY\nos.environ['TAVILY_API_KEY'] = TAVILY_API_KEY<\/code><\/pre>\nSet up OpenAI embeddings<\/h4>\n
from langchain.text_splitter import RecursiveCharacterTextSplitter\nfrom langchain_community.document_loaders import WebBaseLoader\nfrom langchain_community.vectorstores import Chroma\nfrom langchain_openai import OpenAIEmbeddings\nopenai_embed_model = OpenAIEmbeddings(model=\"text-embedding-3-small\")<\/code><\/pre>\nLoad Knowledge Base<\/h4>\n
![\"Workflow](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/Workflow-Graph.webp\")
<\/figure>\n<\/div>\n![\"Harsh](\"https:\/\/av-eks-lekhak.s3.amazonaws.com\/media\/lekhak-profile-images\/converted_image_0fBqNLi.webp\")
<\/p>\n